Sensor arrays with metal-oxide gas sensors for gas detection and determination of gas concentrations in real time

This paper presents a technique for gas analysis in real time by sensor arrays, containing non-linear sensor elements. The methods is based on a non-linear dynamical model. In addition to the concentrations of reducing gases it utilizes also the absolute pressure p(t). The sensor signals of the gas sensors are processed in three steps. First, non-linear filters compensate the distortion in the sensor signals, caused by the sensors. Next, a neural network detects the reducing gases and finally, the gas concentrations are determined. LEOS, a leak detection and location system, will incorporate this developed technique in the future. The technique is verified by measurements of the well defined concentrations of different gases as a function of time. These measurements are also used by the calibration procedure of the gas sensors. The results demonstrate that neural networks unequivocally classify the reducing gases (ketones, alcohols, hydrocarbons, benzole, NH3 and H-2). In addition, the methods is capable of a precise reconstruction of the concentration as a function of time.